Distribution-Free Control Charts Based on Runs and Patterns

TL;DR

This paper introduces distribution-free control charts based on runs and patterns for detecting shifts in data, utilizing permutation properties of Bernoulli trials to control false alarm rates and achieve desired in-control average run lengths.

Contribution

It develops new runs-based control charts with exact conditional distributions, enabling distribution-free monitoring with controlled false alarm probabilities.

Findings

Exact conditional distributions derived for runs statistics.

Control charts achieve specified in-control average run length.

Numerical evaluations demonstrate effective performance.

Abstract

We propose distribution-free runs-based control charts for detecting location shifts. Using the fact that given the number of total successes, the outcomes of a sequence of Bernoulli trials are random permutations, we are able to control the conditional probability of a signal detected at current time given that there is not alarm before at a pre-determined level. This leads to a desired in-control average run length and data-dependent control limits. Two common runs statistics, the longest run statistic and the scan statitsic, are studied in detail and their exact conditional distributions given the number of total successes are obtained using the finite Markov chain imbedding technique. Numerical results are given to evaluate the performance of the proposed control charts.